tests/testthat/test-model-validation.R

In blorr: Tools for Developing Binary Logistic Regression Models

context('test-blr-confusion-matrix')

model <- glm(
  honcomp ~ female + read + science, data = hsb2,
  family = binomial(link = "logit")
)

test_that("prints the correct output", {

  k <- blr_confusion_matrix(model, cutoff = 0.4)

  x <- cat("Confusion Matrix and Statistics

          Reference
Prediction   0   1
         0 125  16
         1  22  37


                Accuracy : 0.8100
     No Information Rate : 0.7350

                   Kappa : 0.5293

McNemars's Test P-Value  : 0.4173

             Sensitivity : 0.6981
             Specificity : 0.8503
          Pos Pred Value : 0.6271
          Neg Pred Value : 0.8865
              Prevalence : 0.2650
          Detection Rate : 0.1850
    Detection Prevalence : 0.2950
       Balanced Accuracy : 0.7742
               Precision : 0.6271
                  Recall : 0.6981

        'Positive' Class : 1")

  expect_output(print(k), x)
})


test_that('throws an error when the model is not of the class glm', {
  model <- lm(read ~ female + science, data = hsb2)
  error <- cat("Hmmm.. Looks like you have specified an incorrect model. The below steps might be helpful:\n\n* Check if you have used the glm() function to build the model.\n* If you have never used it before, you can learn more by typing ?glm or help(glm) in the Console.\n\n Please specify the model in the below format:\n\n glm( honcomp ~ female + read + science , data =  hsb2 , family = binomial(link = 'logit'))\n\n Happy modeling :)")
  expect_error(blr_confusion_matrix(model), error)
})

test_that('cutoff value is between 0 and 1', {
  error <- cat('cutoff can take on values between 0 and 1 only. You have used 1.4 , please specify a value between 0 and 1 only.')
  expect_error(blr_confusion_matrix(model, 1.4), error)
})